Mercury emission proxy mapping at coal and chlor-alkali facilities
No operational satellite retrieves atmospheric mercury directly, but TROPOMI's daily SO2 and NO2 columns expose the combustion and process activity that drives mercury release at coal plants and chlor-alkali facilities.
Sensors
- Sentinel-5P TROPOMI: Primary proxy sensor. Retrieves tropospheric SO2 and NO2 columns at 3.5 × 5.5 km (SO2) and 3.5 × 7 km (NO2) pixel size since 2018, daily global coverage. SO2 detection limit roughly 0.5 DU for strong point sources under favourable conditions; performance degrades significantly under cloud or low solar elevation.
- Aura OMI: Predecessor UV-visible spectrometer, operational since 2004 at 13 × 24 km nadir pixel. Lower spatial resolution than TROPOMI but provides a 20-year archive useful for long-term trend analysis of SO2 and NO2 at persistent industrial sites.
- Landsat-8/9 OLI and TIRS: 30 m multispectral and 100 m thermal infrared imagery at 16-day revisit per satellite. Used to confirm facility operational status, stack activity and cooling-water discharge as independent activity indicators. Cannot detect trace gases but provides spatial context TROPOMI cannot resolve.
- VIIRS Day/Night Band (DNB): Detects facility lighting at roughly 750 m resolution, nightly. Persistent nocturnal illumination at a chlor-alkali or coal plant is a simple, cloud-independent indicator of continuous operation, complementing daytime gas-column retrievals.
Why the satellite cannot see mercury, and why that is not the end of the story
Atmospheric mercury exists primarily as gaseous elemental mercury (GEM) at concentrations measured in nanograms per cubic metre. No current operational satellite spectrometer has the sensitivity or the spectral architecture to retrieve it. The physics is unfavourable: GEM's strongest absorption features sit in the far UV, where solar backscatter is weak and interference from ozone is severe. This is not a solvable problem with incremental instrument improvement; it requires a dedicated mission that does not yet exist.
What satellites can see is the combustion chemistry that accompanies mercury release. Coal combustion emits SO2 in proportion to fuel sulphur content and emits NO2 as a byproduct of high-temperature nitrogen oxidation. Chlor-alkali plants, which use mercury cells to produce chlorine and caustic soda, emit mercury during cell-room ventilation but also generate process heat and ancillary combustion emissions. The published scientific literature treats SO2 column enhancement as a proxy for coal-combustion activity and, by extension, for the mercury flux that co-occurs with it. The inference is probabilistic, not direct.
What the proxy method actually measures, and what it assumes
TROPOMI's SO2 retrieval uses the differential optical absorption spectroscopy (DOAS) method across the 312 to 326 nm UV window. A column enhancement above a known facility, persisting across multiple overpasses, indicates active combustion. Researchers have used this signal to rank facilities by relative emission intensity and to detect curtailment events, for example during Chinese New Year shutdowns or unplanned outages, where SO2 columns drop sharply and then recover.
The proxy assumption is that mercury emission scales with combustion intensity at a given facility, which is broadly true for uncontrolled or partially controlled coal plants. It breaks down when flue-gas desulphurisation (FGD) is fitted: FGD strips SO2 efficiently but has variable and often lower efficiency for mercury. A plant running FGD can show a low SO2 column while still emitting substantial mercury. Chlor-alkali facilities present a different problem: their mercury emissions are largely process-fugitive rather than stack-combustion derived, so the SO2 proxy has weaker physical justification there. NO2 can serve as a secondary indicator of overall plant activity, but it is even less mechanistically coupled to mercury than SO2.
Any analysis must state these assumptions explicitly. The proxy identifies activity and relative intensity; it does not produce a mercury mass flux in tonnes per year without ground-truth emission factors and stack-specific knowledge.
Reading the TROPOMI column: signal, noise and the cloud problem
TROPOMI's SO2 product comes in several vertical column density (VCD) algorithms optimised for different plume heights. For low-altitude industrial sources, the planetary boundary layer (PBL) product is appropriate. Retrievals are flagged by quality value (QA); only pixels with QA above 0.5 are generally used in published facility-level studies. Cloud radiance fraction above roughly 0.3 contaminates or blocks the retrieval entirely.
In practice, a facility in a persistently cloudy region, much of equatorial Asia and central Africa, may have usable retrievals on fewer than 40 percent of days. Seasonal compositing over 30 to 90 days is standard practice to build a stable signal. This means the method is better suited to monitoring chronic emission patterns than to detecting a single day's exceedance. Latency from overpass to publicly available Level-2 product is typically three to five hours for near-real-time files and 24 hours for offline reprocessed data.
Attributing the column to a specific facility
At 3.5 km pixel size, TROPOMI can separate large isolated facilities but struggles where multiple plants cluster within a few tens of kilometres of each other, which is common in the Yangtze River Delta, the Ruhr Valley and parts of India's coal belt. Wind-rotation methods, which correlate column enhancements with concurrent wind fields from ERA5 or similar reanalysis products, help attribute a plume to its source by tracing the enhancement back upwind. This approach has been validated against reported inventories for large SO2 sources in published literature but carries uncertainty that grows with source density and wind variability.
Landsat imagery at 30 m resolution can confirm which stacks within a facility cluster are active on a given overpass date, providing sub-facility attribution that TROPOMI cannot supply alone. The combination is more informative than either sensor alone, though the 16-day Landsat revisit means the pairing is opportunistic rather than systematic.
Honest limits and what the analysis cannot claim
Three limits deserve plain statement. First, the method produces a proxy for mercury-relevant activity, not a mercury measurement. Regulatory enforcement agencies cannot use a TROPOMI SO2 column as a direct substitute for stack monitoring data. Second, facilities with effective SO2 controls can appear clean in the satellite record while remaining significant mercury emitters. Third, chlor-alkali plants are harder to characterise than coal plants because their mercury pathway is not primarily combustion-driven.
The method is most defensible as a screening tool: identifying which facilities in a country or region show anomalous or increasing activity, flagging periods of apparent uncontrolled operation, and prioritising ground-based inspection resources. It is also useful for tracking whether announced emission-reduction commitments correspond to observable changes in column loading over time. That is a meaningful contribution to enforcement, even without a direct mercury retrieval.
Satellize runs this type of proxy analysis on open Sentinel-5P and Landsat archives, combining wind-attribution routines with facility registers to produce ranked facility watchlists. The same analytical infrastructure underpins our Tonga crop-estimation programme, where multi-sensor fusion and careful uncertainty accounting are equally non-negotiable.
Building a monitoring programme around proxy data
A practical enforcement-support programme combines three layers. A daily TROPOMI SO2 and NO2 pull, filtered by QA and cloud fraction, forms the operational backbone. Monthly composites reduce noise and reveal trend direction. Landsat and VIIRS spot-checks, triggered when a TROPOMI anomaly exceeds a threshold, provide spatial confirmation and operational context.
The archive depth matters. TROPOMI data runs from November 2017 to present; OMI extends the record to 2004. Comparing a facility's current column loading against its own historical baseline is more defensible than comparing it against a national average, because it controls for meteorological and seasonal confounders. Any serious programme should define its baseline period, document its QA thresholds and wind-attribution method, and state explicitly what a detected anomaly does and does not prove. That transparency is what makes proxy evidence useful in a regulatory or diplomatic context rather than merely suggestive.
Typical figures
| TROPOMI SO2 pixel size | 3.5 × 5.5 km (nadir, since August 2019 upgrade) |
| TROPOMI NO2 pixel size | 3.5 × 7 km (nadir) |
| Revisit frequency | Daily global coverage (Sentinel-5P); OMI daily at coarser 13 × 24 km |
| SO2 detection limit | Approximately 0.5 DU for strong point sources; degrades under cloud fraction > 0.3 or low solar zenith |
| Product latency | 3–5 hours (near-real-time); ~24 hours (offline reprocessed) |
| Landsat thermal context resolution | 30 m (OLI multispectral), 100 m resampled (TIRS); 16-day revisit per satellite |
| VIIRS DNB resolution | ~750 m; nightly |
| Archive depth | TROPOMI: November 2017–present; OMI: 2004–present; Landsat: 1972–present |
| Spectral bands used | UV 312–326 nm (SO2 DOAS), 405–465 nm (NO2 DOAS), SWIR/TIR for thermal context |
| Cloud screening threshold | QA ≥ 0.5 and cloud radiance fraction < 0.3 (standard published practice) |
Analytics Satellize can run
| Facility activity ranking by SO2 column loading | TROPOMI SO2 VCD monthly composite over registered facility coordinates, ranked by mean enhancement above local background | Ranked facility watchlist as GIS layer and PDF report, updated monthly |
| Anomaly alert on column spike | Daily TROPOMI SO2/NO2 pull with threshold exceedance detection relative to facility-specific 90-day rolling baseline | Automated alert feed (JSON or email) with overpass timestamp, column value and QA flag |
| Wind-attributed plume source assignment | ERA5 wind-rotation back-trajectory overlaid on TROPOMI column enhancement to assign plume to upwind facility | Per-event attribution report with confidence statement and wind uncertainty range |
| Operational status confirmation | Landsat OLI/TIRS thermal anomaly and stack-plume detection at 30 m, triggered by TROPOMI anomaly flag | Annotated Landsat scene with stack activity classification, delivered within 48 hours of suitable overpass |
| Nocturnal activity indicator | VIIRS DNB radiance time series at facility footprint, compared against pre-defined operational illumination baseline | Monthly night-light activity chart per facility, flagging shutdown or curtailment periods |
| Multi-year trend analysis | OMI (2004–2017) spliced with TROPOMI (2017–present) annual SO2 composite trend, bias-corrected for sensor differences using overlap period | Trend report with facility-level time series, suitable for regulatory or diplomatic briefing |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.